Packet-Scheduling and Shared Channel Transmission
In wireless communication systems employing packet-scheduling, at least part of the air-interface resources are assigned dynamically to different users (mobile stations—MS). Those dynamically allocated resources are typically mapped to at least one shared data channel (SDCH). A shared data channel may for example have one of the following configurations:                One or multiple codes in a CDMA (Code Division Multiple Access) system are dynamically shared between multiple MS.        One or multiple subcarriers (subbands) in an OFDMA (Orthogonal Frequency Division Multiple Access) system are dynamically shared between multiple MS.        Combinations of the above in an OFCDMA (Orthogonal Frequency Code Division Multiplex Access) or a MC-CDMA (Multi Carrier-Code Division Multiple Access) system are dynamically shared between multiple MS.        
FIG. 1 shows a resource-scheduling system on a communication channel for systems with e.g. a single shared data channel. A transmission time interval (TTI) reflects the smallest interval at which the scheduler (e.g. the Physical Layer or MAC Layer Scheduler) performs the dynamic resource allocation (DRA). In FIG. 1, a TTI equal to one sub-frame (also referred to as a time slot) is assumed. It should be noted that generally a TTI may also span over multiple sub-frames.
Further, the smallest unit of radio resources (also referred to as a resource block), that can be allocated in OFDMA systems, is typically defined by one TTI in time domain and by one subcarrier/subband in the frequency domain. Similarly, in a CDMA system this smallest unit of radio resources is defined by a TTI in the time domain and a code in the code domain.
In OFCDMA or MC-CDMA systems, this smallest unit is defined by one TTI in time domain, by one subcarrier/subband in the frequency domain and one code in the code domain. Note that dynamic resource allocation may be performed in time domain and in code/frequency domain.
The main benefits of packet-scheduling are the multi-user diversity gain by time domain scheduling (TDS) and dynamic user rate adaptation (DRA).
Assuming that the channel conditions of the users change over time due to fast (and slow) fading, at a given time instant the scheduler can assign available resources (codes in case of CDMA, subcarriers/subbands in case of OFDMA) to users having good channel conditions in time domain scheduling.
Specifics of DRA and Shared Channel Transmission in OFDMA
Additionally to exploiting multi-user diversity in time domain by Time Domain Scheduling (TDS), in OFDMA multi-user diversity can also be exploited in frequency domain by Frequency Domain Scheduling (FDS). This is because the OFDM signal is constructed out of multiple narrowband subcarriers (typically grouped into subbands) in frequency domain, which can be assigned dynamically to different users. By this, the frequency selective channel properties due to multi-path propagation can be exploited to schedule users on frequencies (subcarriers/subbands) on which they have a good channel quality (multi-user diversity in frequency domain).
In an OFDMA system the bandwidth is divided into multiple subbands for practical reasons that consist out of multiple subcarriers. I.e. the smallest unit on which a user may be allocated would have a bandwidth of one subband and a duration of one sub-frame (which may correspond to one or multiple OFDM symbols), which is denoted as a resource block (RB). Typically a subband consists of consecutive subcarriers. However in some case it is desired to form a subband out of distributed non-consecutive subcarriers. A scheduler may also allocate a user over multiple consecutive or non-consecutive subbands and/or sub-frames.
For the 3GPP Long Term Evolution (see 3GPP TR 25.814: “Physical Layer Aspects for Evolved UTRA”, Release 7, v. 7.0.0, June 2006—available at http://www.3gpp.org and incorporated herein by reference), a 10 MHz system may consist out of 600 subcarriers with a subcarrier spacing of 15 kHz. The 600 subcarriers may then be grouped into 24 subbands (a 25 subcarriers), each subband occupying a bandwidth of 375 kHz. Assuming, that a sub-frame has a duration of 0.5 ms, a resource block (RB) would span over 375 kHz and 0.5 ms according to this example.
In order to exploit multi-user diversity and to achieve scheduling gain in frequency domain, the data for a given user should be allocated on resource blocks on which the user has a good channel condition. Typically, those resource blocks are located close to each other and therefore, this transmission mode is also denoted as localized mode (LM). FIG. 2 shows an exemplary data transmission to users in an OFDMA system in localized mode (LM) having a distributed mapping of Layer 1/Layer 2 control signalling.
Alternatively, the users may be allocated in a distributed mode (DM). In this configuration a user (mobile station) is allocated on multiple resource blocks, which are distributed over a range of resource blocks. In distributed mode a number of different implementation options are possible. For exemplary purposes a data transmission to users in an OFDMA system in distributed mode (DM) having a distributed mapping of Layer 1/Layer 2 control signalling is shown in FIG. 3.
Link Adaptation
In mobile communication systems link adaptation is a typical measure to exploit the benefits resulting from dynamic resource allocation. One link adaptation technique is AMC (Adaptive Modulation and Coding). Here, the data-rate per data block or per scheduled user is adapted dynamically to the instantaneous channel quality of the respective allocated resource by dynamically changing the modulation and coding scheme (MCS) in response to the channel conditions. This may require a transmitter to have or obtain a channel quality estimate for the link to the respective receiver. Typically hybrid ARQ (HARQ) techniques are employed in addition. In some configurations it may also make sense to use fast/slow power control.
Channel Quality Information (CQI) Transmission
In a multi-user centrally managed system, a scheduler assigns transmission resources to several users as has been outlined above. Since generally the channel conditions for different users will vary over at least time and frequency, some sort of channel state or channel quality information is required at the scheduler, preferably transmitted from each user equipment device to the scheduler entity.
For most multi-user scheduler algorithms (except Round Robin), the most accurate channel state information should be for the strongest resource blocks, to optimally assign a resource to a user where the channel exhibits a good quality. This will further be used in case that for transmission of data, the modulation or coding scheme is adapted to the channel quality, to increase the spectral efficiency, i.e. in cases where link adaptation is performed.
Generally the CQI is transmitted from a transmitting entity to a receiver entity. In the context of 3G radio network as in UMTS, where a NodeB may act as the multi-user management entity, as well as a multi-cell management entity, the CQI for the downlink transmission chain is obtained (estimated) by a user equipment (UE), which subsequently transmits CQI to a NodeB. Therefore with respect to CQI transmission the user equipment acts as the transmitter entity, and the NodeB as the receiver entity.
Full Feedback
In case a full feedback is transmitted, i.e. the CQI information is not compressed prior to transmission, a CQI value for each of the Nrb resource blocks is transmitted, giving the highest accuracy of information at a very high cost of required transmission bits. To get a rough estimate of the overhead on the CQI feedback information, a system based on the following configurations may be considered: the communication system is equipped with 2×2 MIMO (Multiple Input Multiple Output) using PARC (Per Antenna Rate Control), 20 MHz transmission bandwidth (48 Resource Blocks), 0.5 ms CQI feedback interval, ⅓ rate turbo encoding, no-repetitions or puncturing, and with 24 bit CRC attached. The total CQI feedback overhead of this configuration would be 2.904 Mbps per user.
CQI Compression
One approach to reduce the overhead induced by CQI signalling has been suggested in 3GPP RAN WG#1 Tdoc. R1-061777, “DCT based CQI reporting scheme”, available at http://www.3gpp.org and incorporated herein by reference. The document proposes a scheme using a Discrete Cosine Transform (DCT) to concentrate information into a small number of coefficients and discusses different mechanisms which coefficients to transmit.
Strongest-M DCT and First-M DCT
The “Strongest-M” DCT scheme transmits the DC component of the transformation and in addition M−1 most significant DCT coefficients. Assuming that M is known to transmitter and receiver, only indices of the transmitted coefficients as well as the values of the transmitted coefficients need to be signalled. If M is not known by either the transmitter or the receiver, the value of M may have to be signalled as well.
The “First-M” DCT scheme transmits the M coefficients with the M lowest index values. Assuming that M is known to transmitter and receiver, only the values of the transmitted coefficients need to be signalled. If M is not known by either the transmitter or the receiver, the value of M may have to be signalled as well.
An example of a channel snapshot and an exemplary reconstruction of the channel power using “Strongest 5” DCT scheme is shown in FIG. 8. The corresponding DCT of the complete (“Full DCT”) and compressed (“Strongest 5” DCT) channel information is shown in FIG. 9. While the channel state may be reconstructed perfectly if all DCT coefficients (“Full DCT”) are transmitted, the channel state reconstruction will generally be suboptimum if only a subset of the DCT coefficients is transmitted. The choice of which DCT coefficients are transmitted will affect the accuracy of the reconstructed channel state.
In the “Strongest 5” DCT scheme, only the 5 components with the largest magnitude are chosen in the compression scheme. Since the DC component may be of increased importance, and as it can usually be expected to be among the strongest components anyway, it may be preferable to always transmit the DC coefficient. A bitmap that shows which 5 of the 24 DCT components have the largest magnitude is given in FIG. 10, where a “1” value that the DCT component of that particular index belongs to one of the M largest magnitude coefficients.
It is a matter of convention whether the DCT components are labelled (numbered) from 0 to Nrb−1 or from 1 to Nrb, or similar. Either way usually the DCT component with the lowest index is commonly referred to as the “DC coefficient” or “DC component” (DC=Direct Current). Without loss of generality a numbering ranging from 1 to Nrb is assumed in the examples described herein.
While the above mentioned approaches for transmitting the CQI information are based on performing a DCT on the channel state information and encoding the resulting coefficients, there also exist other schemes where the channel state information, i.e. the individual power levels per resource block are encoded without performing a transformation.
3GPP RAN WG#1 Tdoc. R1-61819, “Overhead reduction of UL CQI signalling for E-UTRA DL”, available at http://www.3gpp.org and incorporated herein by reference, discusses a “Best-M” scheme for feedback reduction of channel quality signalling where a UE reports a label which indicates the M resource blocks with highest signal quality and additionally a single channel quality indicator for these resource blocks. Assuming that M is known to the transmitter and the receiver, signalling of the M selected indices and the selected M values is needed in a CQI report.
A further scheme referred to as “Best M Individual” scheme reports the power for each of the M best resource blocks, and average power for other resource blocks. Assuming that M is known to the transmitter and the receiver, signalling of the M selected indices, the selected M values, and the average value is needed in a CQI report. An exemplary bitmap that signals the best 5 out of 24 resource blocks is shown in FIG. 13.
A further scheme referred to as “Best M Average” reports the average power for M best resource blocks, and average power for other resource blocks. Assuming that M is known to the transmitter and the receiver, signalling of the M selected indices and the two average values is needed in a CQI report. An exemplary bitmap that signals the best 5 out of 24 resource blocks is shown in FIG. 13.
An example of a channel snapshot and an exemplary reconstruction of the channel power using a “Best 5 Individual” scheme and a “Best 5 Average” scheme are shown in FIG. 11 and in FIG. 12, respectively. As can be seen, the “Best 5 Individual” scheme manages to give exact information for the 5 strongest resource blocks (number 8, 9, 10, 18, 19), but quite substantial deviations from the correct value for all other resource blocks. The “Best 5 Average” scheme gives by chance quite accurate information for resource blocks 18 and 19, while we can identify larger deviations—both better and worse—from the correct value for resource blocks 8, 9, and 10. Likewise, for all other resource blocks the reconstructed value may exhibit large differences from the correct values.
Average CQI
Another scheme to reduce the CQI values is to determine the average CQI value and transmit this average value. This may be interpreted as a special case of a Best M=Nrb Average or Best M=0 Average scheme. It requires the least amount of transmitted information, however it also offers a generally very low accuracy with respect to the reconstructed resource block-wise channel quality information.
Signalling
Obviously, there is a need for using information symbols to convey the CQI from the transmitter to the receiver. Without loss of generality, it may be assumed that bits can be used as information symbols. Using the notations defined in subsequent sections, the number of bits required for such signalling is illustrated in Table 1.
TABLE 1Number ofCQI Schemerequired bitsFull FeedbackD · NrbAverageD Best M Individual      D    ·          (              M        +        1            )        +      ⌈          ld      ⁡              (                                                            N                rb                                                                        M                                      )              ⌉   Best M Average      2    ·    D    +      ⌈          ld      ⁡              (                                                            N                rb                                                                        M                                      )              ⌉   DCT Greatest M (assuming that DC coefficient is always transmitted)      D    ·    M    +      ⌈          ld      ⁡              (                                                                              N                  rb                                -                1                                                                                        M                -                1                                                    )              ⌉  
As can be calculated from Table 1 and has been indicated above, the full feedback scheme requires a very high amount of bits to signal the CQI. This requirement may be too high to fulfil in a transmission system, particularly in cellular mobile radio systems where a large number of entities have to report CQI values.
Also DCT-based schemes do not offer an optimal solution for transmitting the CQI information. Since only a limited number of coefficients is transmitted in a DCT compression scheme, the reconstruction at the receiver (which typically offers scheduling functions) is generally not optimum for any resource block. Consequently there will be deviations for the strongest resource blocks, which will result in erroneous scheduler decisions or suboptimum adaptive modulation and coding decisions by the link adaptation entity. Consequently the spectral efficiency is reduced.
In the “Best M Individual” scheme, very detailed information on the channel state is transmitted for the strongest M resource blocks. For all other resource blocks, the information available at the scheduler is extremely rudimentary.
Particularly in case that M is rather small, a problem occurs if a user is assigned more resource blocks than M resource blocks. In this case, some allocated resources are only allocated according to an average resource block quality, which certainly is suboptimum. Furthermore, a subsequent link adaptation would also be based on such an average value, resulting in suboptimum link adaptation and consequently in reduced spectral efficiency. This problem may be circumvented by a high number M, however at the drawback that a lot of feedback signalling is required in this case. Therefore another potential problem is to suggest a coding scheme that requires a small amount of feedback signalling.
In the “Best M Average” scheme, the problems are two-fold. On the one hand, a small number of M will result in similar problems as a small M in the “Best M Individual” scheme. Additionally, the accuracy of the best M resource blocks reported is not as high as in the “Best M Individual” scheme, further deteriorating the accuracy of scheduling or link adaptation performance.
On the other hand, a simple increase of M is not guaranteed to improve the behaviour of the “Best M Average” scheme. Even though the number of resource blocks which are contained within the signalled set increases, the averaging over those M resources will decrease the accuracy for those resource blocks. Therefore there is an optimum M for which the number and level of detail provide the most accurate allocation or link adaptation.
In any case, finding this value of M may not be trivial in a mobile or cellular environment, and—in addition—even when having found an appropriate M value, the achievable data transmission throughput in data transmission is generally bad because of the averaging feature of this scheme.
It should be obvious to those skilled in the art that the information conveyed by the average CQI scheme is of very low accuracy. In order to perform meaningful resource scheduling or link adaptation using CQI-dependant modulation or coding schemes, a higher accuracy than that provided by the average scheme has to be available.